crosstyan/RapidPoseTriangulation
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Export pose postprocessing steps as well.

Browse Source
This commit is contained in:
Daniel
2024-11-29 17:59:15 +01:00
parent 19015c6326
commit ad1ef42b99
3 changed files with 7 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
extras/easypose/pose.py
View File

@ -50,14 +50,8 @@ class SimCC(BaseModel):
return tensor return tensor
def postprocess(self, tensor: List[np.ndarray]): def postprocess(self, tensor: List[np.ndarray]):
simcc_x, simcc_y = tensor kpts = tensor[0][0]
simcc_x = np.squeeze(simcc_x, axis=0) scores = np.expand_dims(tensor[1][0], axis=-1)
simcc_y = np.squeeze(simcc_y, axis=0) keypoints = np.concatenate([kpts, scores], axis=-1)
keypoints = simcc_decoder(simcc_x,
simcc_y,
self.input_shape[2:],
self.dx,
self.dy,
self.scale)
return keypoints return keypoints

4
extras/mmdeploy/configs/pose-detection_simcc_onnxruntime_static-384x288.py
View File

@ -2,7 +2,7 @@ _base_ = ["./pose-detection_static.py", "../_base_/backends/onnxruntime.py"]
onnx_config = dict( onnx_config = dict(
input_shape=[288, 384], input_shape=[288, 384],
output_names=["simcc_x", "simcc_y"], output_names=["kpts", "scores"],
) )
codebase_config = dict(export_postprocess=False) # do not export get_simcc_maximum codebase_config = dict(export_postprocess=True) # export get_simcc_maximum

4
extras/mmdeploy/configs/pose-detection_simcc_onnxruntime_static-384x288_fp16.py
View File

@ -2,7 +2,7 @@ _base_ = ["./pose-detection_static.py", "../_base_/backends/onnxruntime-fp16.py"
onnx_config = dict( onnx_config = dict(
input_shape=[288, 384], input_shape=[288, 384],
output_names=["simcc_x", "simcc_y"], output_names=["kpts", "scores"],
) )
codebase_config = dict(export_postprocess=False) # do not export get_simcc_maximum codebase_config = dict(export_postprocess=True) # export get_simcc_maximum